Add kube_job_status_suspended metric

Author: Indresh2410

docs/metrics/workload/job-metrics.md

@@ -17,3 +17,4 @@
 | kube_job_complete                     | Gauge       |                                                                                                                           | `job_name`=&lt;job-name&gt; <br> `namespace`=&lt;job-namespace&gt; <br> `condition`=&lt;true\|false\|unknown&gt;                                                                                            | STABLE       |
 | kube_job_failed                       | Gauge       |                                                                                                                           | `job_name`=&lt;job-name&gt; <br> `namespace`=&lt;job-namespace&gt; <br> `condition`=&lt;true\|false\|unknown&gt;                                                                                            | STABLE       |
 | kube_job_created                      | Gauge       |                                                                                                                           | `job_name`=&lt;job-name&gt; <br> `namespace`=&lt;job-namespace&gt;                                                                                                                                          | STABLE       |
+| kube_job_status_suspended                      | Gauge       |                                                                                                                           | `job_name`=&lt;job-name&gt; <br> `namespace`=&lt;job-namespace&gt; <br> `condition`=&lt;true\|false\|unknown&gt;                                                                                                                                          | EXPERIMENTAL       |

internal/store/job.go

@@ -355,6 +355,30 @@ func jobMetricFamilies(allowAnnotationsList, allowLabelsList []string) []generat
 				}
 			}),
 		),
+		*generator.NewFamilyGeneratorWithStability(
+			"kube_job_status_suspended",
+			"The number of pods which reached Phase Suspended.",
+			metric.Gauge,
+			basemetrics.ALPHA,
+			"",
+			wrapJobFunc(func(j *v1batch.Job) *metric.Family {
+				ms := []*metric.Metric{}
+				for _, c := range j.Status.Conditions {
+					if c.Type == v1batch.JobSuspended {
+						metrics := addConditionMetrics(c.Status)
+						for _, m := range metrics {
+							metric := m
+							metric.LabelKeys = []string{"condition"}
+							ms = append(ms, metric)
+						}
+					}
+				}
+
+				return &metric.Family{
+					Metrics: ms,
+				}
+			}),
+		),
 		*generator.NewFamilyGeneratorWithStability(
 			"kube_job_owner",
 			"Information about the Job's owner.",

internal/store/job_test.go

@@ -77,7 +77,10 @@ func TestJobStore(t *testing.T) {
 		# HELP kube_job_status_start_time [STABLE] StartTime represents time when the job was acknowledged by the Job Manager.
 		# TYPE kube_job_status_start_time gauge
 		# HELP kube_job_status_succeeded [STABLE] The number of pods which reached Phase Succeeded.
-		# TYPE kube_job_status_succeeded gauge`
+		# TYPE kube_job_status_succeeded gauge
+		# HELP kube_job_status_suspended The number of pods which reached Phase Suspended.
+		# TYPE kube_job_status_suspended gauge
+		`
 
 	cases := []generateMetricsTestCase{
 		{
@@ -272,6 +275,42 @@ func TestJobStore(t *testing.T) {
 				kube_job_status_failed{job_name="SuccessfulJob2NoActiveDeadlineSeconds",namespace="ns1"} 0
 				kube_job_status_start_time{job_name="SuccessfulJob2NoActiveDeadlineSeconds",namespace="ns1"} 1.495800607e+09
 				kube_job_status_succeeded{job_name="SuccessfulJob2NoActiveDeadlineSeconds",namespace="ns1"} 1
+`,
+		},
+		{
+			Obj: &v1batch.Job{
+				ObjectMeta: metav1.ObjectMeta{
+					Name:       "SuspendedNoActiveDeadlineSeconds",
+					Namespace:  "ns1",
+					Generation: 1,
+				},
+				Status: v1batch.JobStatus{
+					Active:    0,
+					Failed:    0,
+					Succeeded: 0,
+					StartTime: &metav1.Time{Time: SuccessfulJob2StartTime},
+					Conditions: []v1batch.JobCondition{
+						{Type: v1batch.JobSuspended, Status: v1.ConditionTrue},
+					},
+				},
+				Spec: v1batch.JobSpec{
+					Suspend:     &trueValue,
+					Parallelism: &Parallelism1,
+					Completions: &Completions1,
+				},
+			},
+			Want: metadata + `
+				kube_job_owner{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1",owner_is_controller="",owner_kind="",owner_name=""} 1
+				kube_job_info{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1
+				kube_job_spec_completions{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1
+				kube_job_spec_parallelism{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1
+				kube_job_status_active{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 0
+				kube_job_status_failed{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 0
+				kube_job_status_start_time{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 1.495800607e+09
+				kube_job_status_succeeded{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1"} 0
+				kube_job_status_suspended{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1",condition="false"} 0
+                kube_job_status_suspended{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1",condition="true"} 1
+				kube_job_status_suspended{job_name="SuspendedNoActiveDeadlineSeconds",namespace="ns1",condition="unknown"} 0
 `,
 		},
 	}